2 * Module for the pnfs nfs4 file layout driver.
3 * Defines all I/O and Policy interface operations, plus code
4 * to register itself with the pNFS client.
7 * The Regents of the University of Michigan
10 * Dean Hildebrand <dhildebz@umich.edu>
12 * Permission is granted to use, copy, create derivative works, and
13 * redistribute this software and such derivative works for any purpose,
14 * so long as the name of the University of Michigan is not used in
15 * any advertising or publicity pertaining to the use or distribution
16 * of this software without specific, written prior authorization. If
17 * the above copyright notice or any other identification of the
18 * University of Michigan is included in any copy of any portion of
19 * this software, then the disclaimer below must also be included.
21 * This software is provided as is, without representation or warranty
22 * of any kind either express or implied, including without limitation
23 * the implied warranties of merchantability, fitness for a particular
24 * purpose, or noninfringement. The Regents of the University of
25 * Michigan shall not be liable for any damages, including special,
26 * indirect, incidental, or consequential damages, with respect to any
27 * claim arising out of or in connection with the use of the software,
28 * even if it has been or is hereafter advised of the possibility of
32 #include <linux/nfs_fs.h>
33 #include <linux/nfs_page.h>
34 #include <linux/module.h>
37 #include "nfs4filelayout.h"
39 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
41 MODULE_LICENSE("GPL");
42 MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
43 MODULE_DESCRIPTION("The NFSv4 file layout driver");
45 #define FILELAYOUT_POLL_RETRY_MAX (15*HZ)
48 filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg,
51 u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count;
55 offset -= flseg->pattern_offset;
56 stripe_no = div_u64(offset, stripe_width);
57 div_u64_rem(offset, flseg->stripe_unit, &rem);
59 return stripe_no * flseg->stripe_unit + rem;
62 /* This function is used by the layout driver to calculate the
63 * offset of the file on the dserver based on whether the
64 * layout type is STRIPE_DENSE or STRIPE_SPARSE
67 filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset)
69 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
71 switch (flseg->stripe_type) {
76 return filelayout_get_dense_offset(flseg, offset);
82 static int filelayout_async_handle_error(struct rpc_task *task,
83 struct nfs4_state *state,
84 struct nfs_client *clp,
87 if (task->tk_status >= 0)
92 switch (task->tk_status) {
93 case -NFS4ERR_BADSESSION:
94 case -NFS4ERR_BADSLOT:
95 case -NFS4ERR_BAD_HIGH_SLOT:
96 case -NFS4ERR_DEADSESSION:
97 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
98 case -NFS4ERR_SEQ_FALSE_RETRY:
99 case -NFS4ERR_SEQ_MISORDERED:
100 dprintk("%s ERROR %d, Reset session. Exchangeid "
101 "flags 0x%x\n", __func__, task->tk_status,
102 clp->cl_exchange_flags);
103 nfs4_schedule_session_recovery(clp->cl_session);
108 rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX);
110 case -NFS4ERR_RETRY_UNCACHED_REP:
113 dprintk("%s DS error. Retry through MDS %d\n", __func__,
122 /* NFS_PROTO call done callback routines */
124 static int filelayout_read_done_cb(struct rpc_task *task,
125 struct nfs_read_data *data)
129 dprintk("%s DS read\n", __func__);
131 if (filelayout_async_handle_error(task, data->args.context->state,
132 data->ds_clp, &reset) == -EAGAIN) {
133 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
134 __func__, data->ds_clp, data->ds_clp->cl_session);
136 pnfs_set_lo_fail(data->lseg);
137 nfs4_reset_read(task, data);
139 rpc_restart_call_prepare(task);
147 * We reference the rpc_cred of the first WRITE that triggers the need for
148 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
149 * rfc5661 is not clear about which credential should be used.
152 filelayout_set_layoutcommit(struct nfs_write_data *wdata)
154 if (FILELAYOUT_LSEG(wdata->lseg)->commit_through_mds ||
155 wdata->res.verf->committed == NFS_FILE_SYNC)
158 pnfs_set_layoutcommit(wdata);
159 dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino,
160 (unsigned long) NFS_I(wdata->inode)->layout->plh_lwb);
164 * Call ops for the async read/write cases
165 * In the case of dense layouts, the offset needs to be reset to its
168 static void filelayout_read_prepare(struct rpc_task *task, void *data)
170 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
172 rdata->read_done_cb = filelayout_read_done_cb;
174 if (nfs41_setup_sequence(rdata->ds_clp->cl_session,
175 &rdata->args.seq_args, &rdata->res.seq_res,
179 rpc_call_start(task);
182 static void filelayout_read_call_done(struct rpc_task *task, void *data)
184 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
186 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
188 /* Note this may cause RPC to be resent */
189 rdata->mds_ops->rpc_call_done(task, data);
192 static void filelayout_read_release(void *data)
194 struct nfs_read_data *rdata = (struct nfs_read_data *)data;
196 rdata->mds_ops->rpc_release(data);
199 static int filelayout_write_done_cb(struct rpc_task *task,
200 struct nfs_write_data *data)
204 if (filelayout_async_handle_error(task, data->args.context->state,
205 data->ds_clp, &reset) == -EAGAIN) {
206 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
207 __func__, data->ds_clp, data->ds_clp->cl_session);
209 pnfs_set_lo_fail(data->lseg);
210 nfs4_reset_write(task, data);
212 rpc_restart_call_prepare(task);
216 filelayout_set_layoutcommit(data);
220 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
221 static void prepare_to_resend_writes(struct nfs_write_data *data)
223 struct nfs_page *first = nfs_list_entry(data->pages.next);
225 data->task.tk_status = 0;
226 memcpy(data->verf.verifier, first->wb_verf.verifier,
227 sizeof(first->wb_verf.verifier));
228 data->verf.verifier[0]++; /* ensure verifier mismatch */
231 static int filelayout_commit_done_cb(struct rpc_task *task,
232 struct nfs_write_data *data)
236 if (filelayout_async_handle_error(task, data->args.context->state,
237 data->ds_clp, &reset) == -EAGAIN) {
238 dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n",
239 __func__, data->ds_clp, data->ds_clp->cl_session);
241 prepare_to_resend_writes(data);
242 pnfs_set_lo_fail(data->lseg);
244 rpc_restart_call_prepare(task);
251 static void filelayout_write_prepare(struct rpc_task *task, void *data)
253 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
255 if (nfs41_setup_sequence(wdata->ds_clp->cl_session,
256 &wdata->args.seq_args, &wdata->res.seq_res,
260 rpc_call_start(task);
263 static void filelayout_write_call_done(struct rpc_task *task, void *data)
265 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
267 /* Note this may cause RPC to be resent */
268 wdata->mds_ops->rpc_call_done(task, data);
271 static void filelayout_write_release(void *data)
273 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
275 wdata->mds_ops->rpc_release(data);
278 static void filelayout_commit_release(void *data)
280 struct nfs_write_data *wdata = (struct nfs_write_data *)data;
282 nfs_commit_release_pages(wdata);
283 if (atomic_dec_and_test(&NFS_I(wdata->inode)->commits_outstanding))
284 nfs_commit_clear_lock(NFS_I(wdata->inode));
285 nfs_commitdata_release(wdata);
288 struct rpc_call_ops filelayout_read_call_ops = {
289 .rpc_call_prepare = filelayout_read_prepare,
290 .rpc_call_done = filelayout_read_call_done,
291 .rpc_release = filelayout_read_release,
294 struct rpc_call_ops filelayout_write_call_ops = {
295 .rpc_call_prepare = filelayout_write_prepare,
296 .rpc_call_done = filelayout_write_call_done,
297 .rpc_release = filelayout_write_release,
300 struct rpc_call_ops filelayout_commit_call_ops = {
301 .rpc_call_prepare = filelayout_write_prepare,
302 .rpc_call_done = filelayout_write_call_done,
303 .rpc_release = filelayout_commit_release,
306 static enum pnfs_try_status
307 filelayout_read_pagelist(struct nfs_read_data *data)
309 struct pnfs_layout_segment *lseg = data->lseg;
310 struct nfs4_pnfs_ds *ds;
311 loff_t offset = data->args.offset;
316 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
317 __func__, data->inode->i_ino,
318 data->args.pgbase, (size_t)data->args.count, offset);
320 if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags))
321 return PNFS_NOT_ATTEMPTED;
323 /* Retrieve the correct rpc_client for the byte range */
324 j = nfs4_fl_calc_j_index(lseg, offset);
325 idx = nfs4_fl_calc_ds_index(lseg, j);
326 ds = nfs4_fl_prepare_ds(lseg, idx);
328 /* Either layout fh index faulty, or ds connect failed */
329 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
330 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
331 return PNFS_NOT_ATTEMPTED;
333 dprintk("%s USE DS: %s\n", __func__, ds->ds_remotestr);
335 /* No multipath support. Use first DS */
336 data->ds_clp = ds->ds_clp;
337 fh = nfs4_fl_select_ds_fh(lseg, j);
341 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
342 data->mds_offset = offset;
344 /* Perform an asynchronous read to ds */
345 status = nfs_initiate_read(data, ds->ds_clp->cl_rpcclient,
346 &filelayout_read_call_ops);
348 return PNFS_ATTEMPTED;
351 /* Perform async writes. */
352 static enum pnfs_try_status
353 filelayout_write_pagelist(struct nfs_write_data *data, int sync)
355 struct pnfs_layout_segment *lseg = data->lseg;
356 struct nfs4_pnfs_ds *ds;
357 loff_t offset = data->args.offset;
362 if (test_bit(NFS_DEVICEID_INVALID, &FILELAYOUT_DEVID_NODE(lseg)->flags))
363 return PNFS_NOT_ATTEMPTED;
365 /* Retrieve the correct rpc_client for the byte range */
366 j = nfs4_fl_calc_j_index(lseg, offset);
367 idx = nfs4_fl_calc_ds_index(lseg, j);
368 ds = nfs4_fl_prepare_ds(lseg, idx);
370 printk(KERN_ERR "NFS: %s: prepare_ds failed, use MDS\n",
372 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
373 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
374 return PNFS_NOT_ATTEMPTED;
376 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s\n", __func__,
377 data->inode->i_ino, sync, (size_t) data->args.count, offset,
380 data->write_done_cb = filelayout_write_done_cb;
381 data->ds_clp = ds->ds_clp;
382 fh = nfs4_fl_select_ds_fh(lseg, j);
386 * Get the file offset on the dserver. Set the write offset to
387 * this offset and save the original offset.
389 data->args.offset = filelayout_get_dserver_offset(lseg, offset);
391 /* Perform an asynchronous write */
392 status = nfs_initiate_write(data, ds->ds_clp->cl_rpcclient,
393 &filelayout_write_call_ops, sync);
395 return PNFS_ATTEMPTED;
399 * filelayout_check_layout()
401 * Make sure layout segment parameters are sane WRT the device.
402 * At this point no generic layer initialization of the lseg has occurred,
403 * and nothing has been added to the layout_hdr cache.
407 filelayout_check_layout(struct pnfs_layout_hdr *lo,
408 struct nfs4_filelayout_segment *fl,
409 struct nfs4_layoutget_res *lgr,
410 struct nfs4_deviceid *id,
413 struct nfs4_deviceid_node *d;
414 struct nfs4_file_layout_dsaddr *dsaddr;
415 int status = -EINVAL;
416 struct nfs_server *nfss = NFS_SERVER(lo->plh_inode);
418 dprintk("--> %s\n", __func__);
420 /* FIXME: remove this check when layout segment support is added */
421 if (lgr->range.offset != 0 ||
422 lgr->range.length != NFS4_MAX_UINT64) {
423 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
428 if (fl->pattern_offset > lgr->range.offset) {
429 dprintk("%s pattern_offset %lld too large\n",
430 __func__, fl->pattern_offset);
434 if (!fl->stripe_unit || fl->stripe_unit % PAGE_SIZE) {
435 dprintk("%s Invalid stripe unit (%u)\n",
436 __func__, fl->stripe_unit);
440 /* find and reference the deviceid */
441 d = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode)->pnfs_curr_ld,
442 NFS_SERVER(lo->plh_inode)->nfs_client, id);
444 dsaddr = get_device_info(lo->plh_inode, id, gfp_flags);
448 dsaddr = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
449 /* Found deviceid is being reaped */
450 if (test_bit(NFS_DEVICEID_INVALID, &dsaddr->id_node.flags))
455 if (fl->first_stripe_index >= dsaddr->stripe_count) {
456 dprintk("%s Bad first_stripe_index %u\n",
457 __func__, fl->first_stripe_index);
461 if ((fl->stripe_type == STRIPE_SPARSE &&
462 fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
463 (fl->stripe_type == STRIPE_DENSE &&
464 fl->num_fh != dsaddr->stripe_count)) {
465 dprintk("%s num_fh %u not valid for given packing\n",
466 __func__, fl->num_fh);
470 if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
471 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
472 "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
478 dprintk("--> %s returns %d\n", __func__, status);
481 nfs4_fl_put_deviceid(dsaddr);
485 static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
489 for (i = 0; i < fl->num_fh; i++) {
490 if (!fl->fh_array[i])
492 kfree(fl->fh_array[i]);
499 _filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
501 filelayout_free_fh_array(fl);
506 filelayout_decode_layout(struct pnfs_layout_hdr *flo,
507 struct nfs4_filelayout_segment *fl,
508 struct nfs4_layoutget_res *lgr,
509 struct nfs4_deviceid *id,
512 struct xdr_stream stream;
514 struct page *scratch;
519 dprintk("%s: set_layout_map Begin\n", __func__);
521 scratch = alloc_page(gfp_flags);
525 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
526 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
528 /* 20 = ufl_util (4), first_stripe_index (4), pattern_offset (8),
530 p = xdr_inline_decode(&stream, NFS4_DEVICEID4_SIZE + 20);
534 memcpy(id, p, sizeof(*id));
535 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
536 nfs4_print_deviceid(id);
538 nfl_util = be32_to_cpup(p++);
539 if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
540 fl->commit_through_mds = 1;
541 if (nfl_util & NFL4_UFLG_DENSE)
542 fl->stripe_type = STRIPE_DENSE;
544 fl->stripe_type = STRIPE_SPARSE;
545 fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
547 fl->first_stripe_index = be32_to_cpup(p++);
548 p = xdr_decode_hyper(p, &fl->pattern_offset);
549 fl->num_fh = be32_to_cpup(p++);
551 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
552 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
555 /* Note that a zero value for num_fh is legal for STRIPE_SPARSE.
556 * Futher checking is done in filelayout_check_layout */
558 max(NFS4_PNFS_MAX_STRIPE_CNT, NFS4_PNFS_MAX_MULTI_CNT))
561 if (fl->num_fh > 0) {
562 fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *),
568 for (i = 0; i < fl->num_fh; i++) {
569 /* Do we want to use a mempool here? */
570 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), gfp_flags);
571 if (!fl->fh_array[i])
574 p = xdr_inline_decode(&stream, 4);
577 fl->fh_array[i]->size = be32_to_cpup(p++);
578 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
579 printk(KERN_ERR "NFS: Too big fh %d received %d\n",
580 i, fl->fh_array[i]->size);
584 p = xdr_inline_decode(&stream, fl->fh_array[i]->size);
587 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
588 dprintk("DEBUG: %s: fh len %d\n", __func__,
589 fl->fh_array[i]->size);
592 __free_page(scratch);
596 filelayout_free_fh_array(fl);
598 __free_page(scratch);
603 filelayout_free_lseg(struct pnfs_layout_segment *lseg)
605 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
607 dprintk("--> %s\n", __func__);
608 nfs4_fl_put_deviceid(fl->dsaddr);
609 kfree(fl->commit_buckets);
610 _filelayout_free_lseg(fl);
613 static struct pnfs_layout_segment *
614 filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
615 struct nfs4_layoutget_res *lgr,
618 struct nfs4_filelayout_segment *fl;
620 struct nfs4_deviceid id;
622 dprintk("--> %s\n", __func__);
623 fl = kzalloc(sizeof(*fl), gfp_flags);
627 rc = filelayout_decode_layout(layoutid, fl, lgr, &id, gfp_flags);
628 if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id, gfp_flags)) {
629 _filelayout_free_lseg(fl);
633 /* This assumes there is only one IOMODE_RW lseg. What
634 * we really want to do is have a layout_hdr level
635 * dictionary of <multipath_list4, fh> keys, each
636 * associated with a struct list_head, populated by calls
637 * to filelayout_write_pagelist().
639 if ((!fl->commit_through_mds) && (lgr->range.iomode == IOMODE_RW)) {
641 int size = (fl->stripe_type == STRIPE_SPARSE) ?
642 fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
644 fl->commit_buckets = kcalloc(size, sizeof(struct list_head), gfp_flags);
645 if (!fl->commit_buckets) {
646 filelayout_free_lseg(&fl->generic_hdr);
649 fl->number_of_buckets = size;
650 for (i = 0; i < size; i++)
651 INIT_LIST_HEAD(&fl->commit_buckets[i]);
653 return &fl->generic_hdr;
657 * filelayout_pg_test(). Called by nfs_can_coalesce_requests()
659 * return true : coalesce page
660 * return false : don't coalesce page
663 filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
664 struct nfs_page *req)
666 u64 p_stripe, r_stripe;
669 if (!pnfs_generic_pg_test(pgio, prev, req) ||
670 !nfs_generic_pg_test(pgio, prev, req))
673 p_stripe = (u64)prev->wb_index << PAGE_CACHE_SHIFT;
674 r_stripe = (u64)req->wb_index << PAGE_CACHE_SHIFT;
675 stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit;
677 do_div(p_stripe, stripe_unit);
678 do_div(r_stripe, stripe_unit);
680 return (p_stripe == r_stripe);
684 filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio,
685 struct nfs_page *req)
687 BUG_ON(pgio->pg_lseg != NULL);
689 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
695 /* If no lseg, fall back to read through mds */
696 if (pgio->pg_lseg == NULL)
697 nfs_pageio_reset_read_mds(pgio);
701 filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio,
702 struct nfs_page *req)
704 BUG_ON(pgio->pg_lseg != NULL);
706 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
712 /* If no lseg, fall back to write through mds */
713 if (pgio->pg_lseg == NULL)
714 nfs_pageio_reset_write_mds(pgio);
717 static const struct nfs_pageio_ops filelayout_pg_read_ops = {
718 .pg_init = filelayout_pg_init_read,
719 .pg_test = filelayout_pg_test,
720 .pg_doio = pnfs_generic_pg_readpages,
723 static const struct nfs_pageio_ops filelayout_pg_write_ops = {
724 .pg_init = filelayout_pg_init_write,
725 .pg_test = filelayout_pg_test,
726 .pg_doio = pnfs_generic_pg_writepages,
729 static bool filelayout_mark_pnfs_commit(struct pnfs_layout_segment *lseg)
731 return !FILELAYOUT_LSEG(lseg)->commit_through_mds;
734 static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
736 if (fl->stripe_type == STRIPE_SPARSE)
737 return nfs4_fl_calc_ds_index(&fl->generic_hdr, j);
742 struct list_head *filelayout_choose_commit_list(struct nfs_page *req)
744 struct pnfs_layout_segment *lseg = req->wb_commit_lseg;
745 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
747 struct list_head *list;
749 /* Note that we are calling nfs4_fl_calc_j_index on each page
750 * that ends up being committed to a data server. An attractive
751 * alternative is to add a field to nfs_write_data and nfs_page
752 * to store the value calculated in filelayout_write_pagelist
753 * and just use that here.
755 j = nfs4_fl_calc_j_index(lseg,
756 (loff_t)req->wb_index << PAGE_CACHE_SHIFT);
757 i = select_bucket_index(fl, j);
758 list = &fl->commit_buckets[i];
759 if (list_empty(list)) {
760 /* Non-empty buckets hold a reference on the lseg */
766 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
768 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
770 if (flseg->stripe_type == STRIPE_SPARSE)
773 return nfs4_fl_calc_ds_index(lseg, i);
776 static struct nfs_fh *
777 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
779 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
781 if (flseg->stripe_type == STRIPE_SPARSE) {
782 if (flseg->num_fh == 1)
784 else if (flseg->num_fh == 0)
785 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
788 return flseg->fh_array[i];
791 static int filelayout_initiate_commit(struct nfs_write_data *data, int how)
793 struct pnfs_layout_segment *lseg = data->lseg;
794 struct nfs4_pnfs_ds *ds;
798 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
799 ds = nfs4_fl_prepare_ds(lseg, idx);
801 printk(KERN_ERR "NFS: %s: prepare_ds failed, use MDS\n",
803 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
804 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
805 prepare_to_resend_writes(data);
806 data->mds_ops->rpc_release(data);
809 dprintk("%s ino %lu, how %d\n", __func__, data->inode->i_ino, how);
810 data->write_done_cb = filelayout_commit_done_cb;
811 data->ds_clp = ds->ds_clp;
812 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
815 return nfs_initiate_commit(data, ds->ds_clp->cl_rpcclient,
816 &filelayout_commit_call_ops, how);
820 * This is only useful while we are using whole file layouts.
822 static struct pnfs_layout_segment *find_only_write_lseg(struct inode *inode)
824 struct pnfs_layout_segment *lseg, *rv = NULL;
826 spin_lock(&inode->i_lock);
827 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
828 if (lseg->pls_range.iomode == IOMODE_RW)
830 spin_unlock(&inode->i_lock);
834 static int alloc_ds_commits(struct inode *inode, struct list_head *list)
836 struct pnfs_layout_segment *lseg;
837 struct nfs4_filelayout_segment *fl;
838 struct nfs_write_data *data;
841 /* Won't need this when non-whole file layout segments are supported
842 * instead we will use a pnfs_layout_hdr structure */
843 lseg = find_only_write_lseg(inode);
846 fl = FILELAYOUT_LSEG(lseg);
847 for (i = 0; i < fl->number_of_buckets; i++) {
848 if (list_empty(&fl->commit_buckets[i]))
850 data = nfs_commitdata_alloc();
853 data->ds_commit_index = i;
855 list_add(&data->pages, list);
861 for (j = i; j < fl->number_of_buckets; j++) {
862 if (list_empty(&fl->commit_buckets[i]))
864 nfs_retry_commit(&fl->commit_buckets[i], lseg);
865 put_lseg(lseg); /* associated with emptying bucket */
868 /* Caller will clean up entries put on list */
872 /* This follows nfs_commit_list pretty closely */
874 filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
877 struct nfs_write_data *data, *tmp;
880 if (!list_empty(mds_pages)) {
881 data = nfs_commitdata_alloc();
885 list_add(&data->pages, &list);
888 if (alloc_ds_commits(inode, &list))
891 list_for_each_entry_safe(data, tmp, &list, pages) {
892 list_del_init(&data->pages);
893 atomic_inc(&NFS_I(inode)->commits_outstanding);
895 nfs_init_commit(data, mds_pages, NULL);
896 nfs_initiate_commit(data, NFS_CLIENT(inode),
899 nfs_init_commit(data, &FILELAYOUT_LSEG(data->lseg)->commit_buckets[data->ds_commit_index], data->lseg);
900 filelayout_initiate_commit(data, how);
905 list_for_each_entry_safe(data, tmp, &list, pages) {
906 nfs_retry_commit(&data->pages, data->lseg);
907 list_del_init(&data->pages);
908 nfs_commit_free(data);
910 nfs_retry_commit(mds_pages, NULL);
911 nfs_commit_clear_lock(NFS_I(inode));
916 filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
918 nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
921 static struct pnfs_layoutdriver_type filelayout_type = {
922 .id = LAYOUT_NFSV4_1_FILES,
923 .name = "LAYOUT_NFSV4_1_FILES",
924 .owner = THIS_MODULE,
925 .alloc_lseg = filelayout_alloc_lseg,
926 .free_lseg = filelayout_free_lseg,
927 .pg_read_ops = &filelayout_pg_read_ops,
928 .pg_write_ops = &filelayout_pg_write_ops,
929 .mark_pnfs_commit = filelayout_mark_pnfs_commit,
930 .choose_commit_list = filelayout_choose_commit_list,
931 .commit_pagelist = filelayout_commit_pagelist,
932 .read_pagelist = filelayout_read_pagelist,
933 .write_pagelist = filelayout_write_pagelist,
934 .free_deviceid_node = filelayout_free_deveiceid_node,
937 static int __init nfs4filelayout_init(void)
939 printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
941 return pnfs_register_layoutdriver(&filelayout_type);
944 static void __exit nfs4filelayout_exit(void)
946 printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
948 pnfs_unregister_layoutdriver(&filelayout_type);
951 MODULE_ALIAS("nfs-layouttype4-1");
953 module_init(nfs4filelayout_init);
954 module_exit(nfs4filelayout_exit);